jev-drone
Camera-only MuJoCo quadrotor where Jev makes 2.5 Hz tactical judgments while code keeps flight control and a safety veto.
Overview
jev-drone flies a simulated Skydio X2 quadrotor through a five-station obstacle course in MuJoCo using only its onboard camera. A layered architecture keeps time-critical work in code: a 500 Hz geometric controller, a 50 Hz guidance and safety reflex that always owns safety, and a 15 Hz camera pipeline that turns depth and segmentation buffers into a compact symbolic scene. Jev runs at about 2.5 Hz as an advisory tactical layer.
In one call Jev answers three questions: which manoeuvre to take (hold course, gap left or right, climb, brake, reacquire), how risky the situation is, and whether the target is truly lost or briefly occluded. Code decides when to ask, skips calls in open corridors, and reuses judgements for fingerprinted unchanged scenes; a typical 65-second flight makes about 110 calls. Code keeps a veto, for example refusing a climb unless the obstacle's measured top edge is within the climb ceiling.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which manoeuvre should the drone take? | hold_course · gap_left · gap_right · climb · brake · reacquire |
| Score | How risky is the current situation? | — |
| Yes / no | Is the target genuinely lost rather than briefly occluded? | — |
How it works
- Camera buffers become a symbolic scene at 15 Hz
- Code decides whether a judgement is needed or can be reused
- Jev answers manoeuvre, risk and target-lost in one call
- Guidance applies advice subject to a 50 Hz safety reflex veto
Strengths
- Safety stays deterministic
- Calls skipped or cached when nothing changes
- Real airframe model, no scripted flight
Caveats
- Simulation only
- Advisory role; not a perception or control model
At a glance
- Type
- Project
- Status
- prototype
- Language
- Python
- Runtime
- MuJoCo simulation
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Robotics simulation: camera-only autonomous drone in MuJoCo that puts a Jev judgment model in the control loop at 2.5 Hz.
yibie/awesome-jevMuJoCo quadrotor: control and safety stay in code; Jev handles slower tactical judgments.
AnotiaWang/awesome-decision-modelsA MuJoCo drone experiment deriving scene features from camera buffers for Jev tactical advice.
logicrw/awesome-jev-projectsCamera-only autonomous drone in MuJoCo with a small judgment model.
wh000wh000/awesome-jev-liveSimulated MuJoCo quadrotor with Jev making slower tactical judgments from processed camera observations; deterministic code controls flight.
cobanov/awesome-jevCamera-only autonomous drone in MuJoCo with a small judgment model (TypeSafe Jev) in the loop at 2.5Hz.
hellogumbo/awesome-jevCamera-only quadrotor in MuJoCo with Jev making judgment calls at about 2.5 Hz.
MrJev/awesome-jev
Frequently asked questions
What does Jev decide?
The tactical manoeuvre, situation risk, and whether the target is truly lost.
Can Jev crash the drone?
Code keeps a veto and a 50 Hz safety reflex overrides any judgement when something is close.
How many calls per flight?
About 110 for a typical 65-second flight, per the README.
Related projects
Simulated Franka arm that follows plain-English goals by having Jev chain about 30 hardcoded motion primitives, one per step.
Agent skillprototypePythonMulti-drone simulation where Jev makes fast reflex decisions per drone and an optional slower LLM may advise but never takes control.
AppprototypeTypeScriptOpen 4B System I driving model that scores behaviours like cruise, stop, yield and brake from three camera frames each decision.
ProjectresearchPythonExperimental robot loop where Jev chooses the next verified action from perception-generated options for a language goal.
ProjectresearchPythonMuJoCo PiPER arm controlled by a local decider-2b model choosing among ten skills via a Jev-style typed-choice interface while targets move.
Agent skillresearchPythonMuJoCo pilot where Jev, Claude Haiku and reactive rules choose pick-and-place actions; Jev matched the rules exactly.
ProjectresearchPython